Weakly supervised perivascular spaces segmentation with salient guidance of Frangi filter

Abstract

To develop a weakly supervised 3D perivascular spaces (PVS) segmentation model that combines the filter-based image processing algorithm and the convolutional neural network. Methods: We present a weakly supervised learning method for PVS segmentation by combing a rule-based image processing approach Frangi filter with a canonical deep learning algorithm Unet using conditional random field theory. The weighted cross entropy loss function and the training patch selection were implemented for the optimization and to alleviate the class imbalance issue. The performance of the model was evaluated on the Human Connectome Project data. Results: The proposed method increases the true positive rate compared to the rule-based method and reduces the false positive rate by 36% in the weakly supervised training experiment and 39.4% in the supervised training experiment compared to Unet, which results in superior overall performance. In addition, by training the model on manually quality controlled and annotated data which includes the subjects with the presence of white matter hyperintensities, the proposed method differentiates between PVS and white matter hyperintensities, which reduces the false positive rate by 78.5% compared to weakly supervised trained model. Conclusions: Combing the filter-based image processing algorithm and the convolutional neural network algorithm could improve the model's segmentation accuracy, while reducing the training dependence on the large scale annotated PVS mask data by the trained physician. Compared to the filter-based image processing algorithm, the data driven PVS segmentation model using quality-controlled data as the training target could differentiate the white matter hyperintensity from PVS resulting low false positive rate.

A Systematic Review and Meta-Analysis of Automated Methods for Quantifying Enlarged Perivascular Spaces in the Brain

“…Gonzalez et al (2016a, 2016b, 2017), Wang et al (2016), and Ballerini et al (2018) all use scans from the Edinburgh Mild Stroke Studies. Other common data sources used are the Human Connectome Project (Shepehrband et al, 2019; Choi et al, 2020; Lan et al, 2023), the Lothian Birth Cohort 1936 (Ballerini et al, 2016, Wang et al, 2016), the Multi-Ethnic Study of Atherosclerosis (MESA) cohort (Rashid et al, 2023) and the Southall And BRent REvisited (SABRE) (Sudre et al, 2022, 4 participating studies). All these are well-known clinical and population studies with data available either by request to the data holders or through public databases, which facilitates comparability in the results.…”

“…Figure 3 comparatively illustrates the distribution of sample sizes of the studies in this category. Dubost et al (2017Dubost et al ( , 2019aDubost et al ( , 2019bDubost et al ( , 2019cDubost et al ( , 2020 (Shepehrband et al, 2019;Choi et al, 2020;Lan et al, 2023), the Lothian Birth Cohort 1936 (Ballerini et al, 2016, the Multi-Ethnic Study of Atherosclerosis (MESA) cohort (Rashid et al, 2023) and the Southall And BRent REvisited (SABRE) (Sudre et al, 2022, 4 participating studies). All these are well-known clinical and population studies with data available either by request to the data holders or through public databases, which facilitates comparability in the results.…”

Parenchymal CSF fraction is a measure of brain glymphatic clearance and positively associated with amyloid beta deposition on PET

“…There are other developed non‐invasive measures for estimating glymphatic function in humans other than the CSFF reported in this work, including the PVS load by segmentation and DTI‐ALPS. PVS segmentation can be performed on T2w but is only feasible for significantly enlarged ones in WM and the segmentation results vary across techniques and image quality 15,43 . DTI‐ALPS is a method that computes fluid diffusivity along PVS, which potentially reflects the glymphatic function 23,31,56,57 .…”

“… 2 , 4 , 38 Varied research outcomes have linked PVS load to Aβ deposition, yet inconsistencies may arise from measurement sensitivity. 14 , 16 , 39 PVS enlargement may reflect CSF fluid stasis and reduced flow leading to overall deficits in glymphatic clearance 40 , 41 , 42 because PVS load segmentation depends on image (typically T2w) resolution and quality, 15 , 43 , 44 and only includes the visibly enlarged PVS surrounding periarteries in WM. 45 , 46 Our CSFF potentially overcomes these limitations by accounting for all parenchymal CSF water in PVS, including microscale PVS that traditional imaging might miss.…”

Deep learning applications in vascular dementia using neuroimaging

“…Research has indicated that enlarged PVS, obtained from a deep learning model, can be linked to the vascular amyloid-β accumulation [37]. CNN models have been showing promising results in the quantification and segmentation of PVS [38]. In addition, CNN models can also quantify PVS in four different brain regions: midbrain, hippocampi, basal ganglia and centrum semiovale [39], and these region-specific PVS can be influenced by sex, cardiovascular risk factors, APOE gene and imaging biomarkers [40 ▪▪ ].…”